With fixed references, three-region guides lowered mean FIP-MLE error in iEEG and fMRI; a shared benefit test failed

We tested whether three other brain regions improve the prediction of local temporal-persistence estimates. Public source tables came from separate normal-tissue iEEG and healthy fMRI studies. Training used 48 electrical patients and 26 controls. Fixed tests used 18 other patients and 26 other controls. The baseline controlled location, hemisphere, and source prefix. We computed exact average error over every allowed three-region guide panel. With fixed references, FIP-MLE mean normalized error decreases were 0.70664 and 0.13056. Pointwise 95% person intervals were 0.24455 to 1.16874 and 0.03916 to 0.22195. Positive scores occurred in 12/18 and 16/26 people. The fixed shared mean-and-majority rule failed, Holm p = 0.32694. A second measure, WPLM, had uncertain fMRI evidence and failed, Holm p = 0.57749. No electrical reference-person resample retained every original panel in 399 draws. Thus no complete reference-resampled electrical mean interval was available. Planned anatomical and processing checks, added demographic checks, missing values, and all negative scores remain. An exact Gaussian example shows that a mean-optimal guide can increase conditional prediction error for most people. A common measurement offset is sufficient; no neural cause is identified. Source values were checked against the pinned original archive. The package gives inputs, provenance, code, local locks, all results, figures, and repeat instructions. This is a new reserved endpoint in reused studies, not independent acquisition confirmation or treatment validation.

intracranial EEG · fMRI · Hurst exponent · normative modeling · prediction · reference uncertainty · person-level inference